Skip to content

Local UI to run and train LLMs and diffusion models. Supports GGUF, MLX, Qwen3.8, DeepSeek-V4.1, MiniMax-H3, Gemma 4, FLUX and more.

Studio: train Phi-4-mini on its own chat template (#17312)

Nilay5ab7b904dfce
diff --git a/studio/backend/tests/test_phi4_mini_own_chat_template.py b/studio/backend/tests/test_phi4_mini_own_chat_template.py
new file mode 100644
index 00000000000..a5380929cf5
--- /dev/null
+++ b/studio/backend/tests/test_phi4_mini_own_chat_template.py
@@ -0,0 +1,70 @@
+# SPDX-License-Identifier: AGPL-3.0-only
+# Copyright 2026-present the Unsloth AI Inc. team. All rights reserved. See /studio/LICENSE.AGPL-3.0
+
+import pytest
+from datasets import Dataset
+
+from utils.datasets import apply_chat_template_to_dataset
+from utils.datasets.completion_masking import lookup_manual_markers
+from utils.datasets.model_mappings import MODEL_TO_TEMPLATE_MAPPER
+
+_PHI4_MINI_TEMPLATE = (
+ "{% for message in messages %}{{ '<|' + message['role'] + '|>' + message['content'] + '<|end|>' }}"
+ "{% endfor %}{% if add_generation_prompt %}{{ '<|assistant|>' }}{% else %}{{ eos_token }}{% endif %}"
+)
+
+_CONVERSATION = [
+ {"role": "user", "content": "What is 2+2?"},
+ {"role": "assistant", "content": "4"},
+]
+
+
+def _tokenizer():
+ from tokenizers import Tokenizer, models, pre_tokenizers
+ from transformers import PreTrainedTokenizerFast
+
+ vocab = {"[UNK]": 0, "<|endoftext|>": 1, "<|user|>": 2, "<|assistant|>": 3, "<|end|>": 4}
+ backend = Tokenizer(models.WordLevel(vocab, unk_token = "[UNK]"))
+ backend.pre_tokenizer = pre_tokenizers.Whitespace()
+ tokenizer = PreTrainedTokenizerFast(
+ tokenizer_object = backend,
+ unk_token = "[UNK]",
+ eos_token = "<|endoftext|>",
+ pad_token = "<|endoftext|>",
+ )
+ tokenizer.chat_template = _PHI4_MINI_TEMPLATE
+ return tokenizer
+
+
+ "name",
+ [
+ "unsloth/Phi-4-mini-instruct",
+ "microsoft/Phi-4-mini-instruct",
+ "unsloth/Phi-4-mini-instruct-unsloth-bnb-4bit",
+ "unsloth/phi-4-mini-reasoning",
+ "microsoft/Phi-4-mini-reasoning",
+ ],
+)
+def test_phi4_mini_trains_on_its_own_chat_template(name):
+ assert lookup_manual_markers(name) == (None, None, None)
+ tokenizer = _tokenizer()
+ expected = tokenizer.apply_chat_template(_CONVERSATION, tokenize = False)
+ result = apply_chat_template_to_dataset(
+ {
+ "dataset": Dataset.from_dict({"messages": [_CONVERSATION]}),
+ "final_format": "chatml_messages",
+ "chat_column": "messages",
+ "is_standardized": True,
+ },
+ tokenizer,
+ model_name = name,
+ num_proc = 1,
+ )
+ assert result["success"], result.get("errors")
+ assert result["dataset"][0]["text"] == expected
+
+
+def test_phi4_keeps_the_phi4_template():
+ assert MODEL_TO_TEMPLATE_MAPPER["unsloth/phi-4"] == "phi-4"
+ assert MODEL_TO_TEMPLATE_MAPPER["unsloth/phi-4-reasoning"] == "phi-4"
diff --git a/studio/backend/utils/datasets/model_mappings.py b/studio/backend/utils/datasets/model_mappings.py
index 84ec18830ba..a5528396f7b 100644
--- a/studio/backend/utils/datasets/model_mappings.py
+++ b/studio/backend/utils/datasets/model_mappings.py
@@ -32,14 +32,6 @@ TEMPLATE_TO_MODEL_MAPPER = {
"unsloth/phi-4-reasoning-plus",
"microsoft/Phi-4-reasoning-plus",
"unsloth/phi-4-reasoning-plus-bnb-4bit",
- "unsloth/phi-4-mini-reasoning-unsloth-bnb-4bit",
- "unsloth/phi-4-mini-reasoning",
- "microsoft/Phi-4-mini-reasoning",
- "unsloth/phi-4-mini-reasoning-bnb-4bit",
- "unsloth/Phi-4-mini-instruct-unsloth-bnb-4bit",
- "unsloth/Phi-4-mini-instruct",
- "microsoft/Phi-4-mini-instruct",
- "unsloth/Phi-4-mini-instruct-bnb-4bit",
),
"mistral": (
"unsloth/mistral-7b-instruct-v0.1-bnb-4bit",